Sorry to be this person, but I don't really agree with the first sentence:
--> "Civic institutions—the rule of law, universities, and a free press—are the backbone of democratic life."
People are the backbone of our civilization. People who have good intentions and support one another. We don't NEED an FDA to function -- it's just a tool that has worked quite well for a long time for us.
There are a lot of tools to address the problem that some people have bad intentions.
We publish common sense laws, and we have police officers and prosecutors, and then we have a court system to hold people accountable for breaking the law. That's one pretty major method that has little to do with the need for an institution like FDA.
I don't know if a system that relied entirely on tort and negligence and contract law to protect people from being sold snake oil would function better or worse than FDA, but I do know something like FDA (where a bunch of smart people advise very specifically on which drugs are ok to take and which are not) isn't the only option we have.
as a lawyer, I'll just note that the legal system has standards for this. specifically, you can't delete stuff that poses legal risks to you once you reasonably expect a lawsuit about it. but you can delete it as part of normal business activity until that point.
Why do we need long running agents? Most of my experienced value with LLMs has been like 1 to 10 turn chats. Should they just ban longer chats to solve these issues?
Because you get the biggest time-savings when you can let it run longer between each time it needs a human in the loop.
I have multi-week runs of Claude Code going to work on a compiler project. I have a week-long run of Claude Code where it is writing a real-time strategy game.
In both cases I occasionally review code, and complain a bit about things it has gotten wrong until it's back on track. In both cases it is working to specs that have produced plans that have produced TODO lists. In the latter it wrote the specs itself. In the former, the specs are externally imposed (rubyspecs test suite).
In both cases it means I get involved ranging from ever tens of minutes to every few hours, but mostly then to just confirm it can continue, with more detailed reviews every day or so.
Having to review output and give instructions every turn would drastically diminish the value.
I'm on the top Claude Max tier. Just upgraded. I could probably make do with the lower one, but I hit the limit (for the first time) on the lower Claude Max this week, and I get enough value from it that I was not about to wait for a session reset.
Protecting users in the bargains we strike with big tech is a worthwhile and noble effort, but privacy law has generally woefully failed to do this.
Millions upon millions have been spent on cookie banners -- people are still arguing about them in this thread -- but there is almost zero benefit to this expense.
The main thing that's good about this, IMO, is that fundamentally training a large language model and privacy law as it's written today cannot coexist. They are incompatible. And allowing someone to break the law forever (as is happening today) is not a good long-term solution.
> Training a large language model and privacy law as it's written today cannot coexist
If they aren't compatible, then the conclusion is abundantly obvious; the LLM has to go, not privacy. Small and questionable economic utility in exchange for a pillar of stable democratic society are NOT negotiable tradeoff.
There is enough data on the internet to train LLMs without breaking a single privacy law. If the economic value of LLMs are as real as the companies like to claim, there is enough data on the internet to train LLMs while paying for proper royalty for every single word.
I don't argue that privacy laws have been perfect. Only a fraction of GDPR seems to actually do much. But bending over backwards because big tech slips a few dollars in the pocket of Brussels is NOT the reason we should revise those laws.
reply